Ras activation by SOS: Allosteric regulation by altered fluctuation

dynamics
Lars Iversen et al.
Science 345, 50 (2014);
DOI: 10.1126/science.1250373

This copy is for your personal, non-commercial use only.

If you wish to distribute this article to others, you can order high-quality copies for your
colleagues, clients, or customers by clicking here.

Downloaded from www.sciencemag.org on July 5, 2014

Permission to republish or repurpose articles or portions of articles can be obtained by
following the guidelines here.

The following resources related to this article are available online at

www.sciencemag.org (this information is current as of July 5, 2014 ):

Updated information and services, including high-resolution figures, can be found in the online
version of this article at:
http://www.sciencemag.org/content/345/6192/50.full.html
Supporting Online Material can be found at:
http://www.sciencemag.org/content/suppl/2014/07/02/345.6192.50.DC1.html
This article cites 68 articles, 17 of which can be accessed free:
http://www.sciencemag.org/content/345/6192/50.full.html#ref-list-1
This article appears in the following subject collections:
Cell Biology
http://www.sciencemag.org/cgi/collection/cell_biol

Science (print ISSN 0036-8075; online ISSN 1095-9203) is published weekly, except the last week in December, by the
American Association for the Advancement of Science, 1200 New York Avenue NW, Washington, DC 20005. Copyright
2014 by the American Association for the Advancement of Science; all rights reserved. The title Science is a
registered trademark of AAAS.
R ES E A RC H | R E S EA R C H A R T I C LE S

MOLECULAR KINETICS (DH), and Pleckstrin homology (PH) domains

(10). Autoinhibition is thought to occur through
steric occlusion of the allosteric site, which can
Ras activation by SOS: Allosteric be released by interaction of the PH domain with
phosphatidylinositol-4,5-bisphosphate (PIP2) or
regulation by altered other negative lipids on the membrane (11, 12).
Several SOS mutations, including R552G in the

fluctuation dynamics helical linker, lead to weakened autoinhibition,

excessive Ras activation, and Noonan Syndrome
developmental disorders (13). (Single-letter ab-
Lars Iversen,1*† Hsiung-Lin Tu,1* Wan-Chen Lin,1 Sune M. Christensen,1† breviations for the amino acid residues are as
Steven M. Abel,2‡ Jeff Iwig,3 Hung-Jen Wu,1§ Jodi Gureasko,3‖ Christopher Rhodes,4 follows: A, Ala; C, Cys; D, Asp; E, Glu; F, Phe; G,
Rebecca S. Petit,1 Scott D. Hansen,1 Peter Thill,5 Cheng-Han Yu,6¶ Dimitrios Stamou,7 Gly; H, His; I, Ile; K, Lys; L, Leu; M, Met; N,
Arup K. Chakraborty,2,5,8,9,10,11 John Kuriyan,1,3,12 Jay T. Groves1,6,12,13# Asn; P, Pro; Q, Gln; R, Arg; S, Ser; T, Thr; V, Val;
W, Trp; and Y, Tyr. In the mutants, other amino
Activation of the small guanosine triphosphatase H-Ras by the exchange factor Son of acids were substituted at certain locations; for
Sevenless (SOS) is an important hub for signal transduction. Multiple layers of regulation, example, R552G indicates that arginine at po-
through protein and membrane interactions, govern activity of SOS. We characterized sition 552 was replaced by glycine.)
the specific activity of individual SOS molecules catalyzing nucleotide exchange in Most biological and biochemical studies of
H-Ras. Single-molecule kinetic traces revealed that SOS samples a broad distribution SOS activity have relied on bulk assays. Physi-
of turnover rates through stochastic fluctuations between distinct, long-lived (more cally distinct aspects of SOS regulation, such as
than 100 seconds), functional states. The expected allosteric activation of SOS by membrane recruitment and allosteric modula-
Ras–guanosine triphosphate (GTP) was conspicuously absent in the mean rate. tion of specific catalytic activity, are intrinsically
However, fluctuations into highly active states were modulated by Ras-GTP. This reveals convolved in such observations (supplementary
a mechanism in which functional output may be determined by the dynamical spectrum text S1). Furthermore, any stochastic variation
of rates sampled by a small number of enzymes, rather than the ensemble average. among SOS molecules, such as fluctuations be-

C
ellular membranes organize signal trans-
duction, serving as platforms for protein
interactions as well as direct modulators
of enzymatic function (1, 2). The activa-
tion of lipid-anchored guanosine triphos-
phatases (GTPases) of the Ras superfamily by
cytosolic guanine nucleotide exchange factors
(GEFs) represents a broadly important class of
membrane-localized signaling reactions. Con-
trol of GEF activity is multilayered and involves
membrane recruitment, lateral interactions on
the membrane surface, as well as allosteric reg-
ulation (3, 4). A recurring feature across several
1
Howard Hughes Medical Institute, Department of Chemistry,
University of California, Berkeley, Berkeley, CA 94720, USA.
2
Department of Chemical Engineering, Massachusetts
Institute of Technology (MIT), Cambridge, MA 02139, USA.
3
Howard Hughes Medical Institute, Department of Molecular
and Cell Biology, University of California, Berkeley, Berkeley,
CA 94720, USA. 4Department of Mechanical Engineering,
University of California, Berkeley, Berkeley, CA 94720, USA.
5
Department of Chemistry, MIT, Cambridge, MA 02139, USA.
6
Mechanobiology Institute, National University of Singapore,
Singapore. 7Department of Chemistry and Nano-Science
Center, University of Copenhagen, Copenhagen, Denmark.
8
Department of Biological Engineering, MIT, Cambridge, MA
02139, USA. 9Ragon Institute of Massachusetts General
Hospital, MIT, and Harvard, Cambridge, MA 02139, USA.
10
Department of Physics, MIT, Cambridge, MA 02139, USA.
11
Institute for Medical Engineering and Science, MIT,
Cambridge, MA 02139, USA. 12Physical Biosciences and
Materials Sciences Divisions, Lawrence Berkeley National
Laboratory, Berkeley, CA 94720, USA. 13Berkeley Education
Alliance for Research in Singapore, 1 Create Way, CREATE
tower level 11, University Town, Singapore 138602.
*These authors contributed equally to this work and are listed in
alphabetical order. †Present address: Department of Chemistry,
University of Copenhagen, Copenhagen, Denmark. ‡Present
address: Department of Chemical and Biomolecular Engineering,
University of Tennessee, Knoxville, TN 37996, USA. §Present
address: Department of Chemical Engineering, Texas A&M
University, College Station, TX 77843, USA. ||Present address:
Epizyme, 400 Technology Square, Cambridge, MA 02139, USA.
¶Present address: Department of Anatomy, The University of Hong
Kong, Hong Kong. #Corresponding author. E-mail: jtgroves@lbl.gov
Ras, Rho, and Arf GTPase subfamily GEFs is the
release of autoinhibition mediated by GTPase
and membrane binding. The molecular mech-
anisms underlying these regulatory couplings
remain poorly understood, in large part because
of the intrinsic experimental challenge of work-
ing within a membrane environment. We recon-
stituted the inner-leaflet signaling geometry of
Son of Sevenless (SOS)–catalyzed nucleotide ex-
change in H-Ras on supported membranes that
were partitioned into arrays of two-dimensional
corrals by lithographically defined chromium dif-
fusion barriers. With this system, we monitored
the real-time catalytic activity of individual SOS
molecules in order to observe the functional
mechanisms of allosteric activation and auto-
inhibition on the membrane surface.
SOS is widely distributed in mammalian cells.
In vivo, inactive cytosolic SOS is recruited to the
plasma membrane in response to ligand bind-
ing by receptors on the cell surface. There, it
activates membrane-tethered Ras by catalyzing
the exchange of Ras-bound guanosine diphos-
phate (GDP) with guanosine triphosphate (GTP),
which triggers the mitogen-activated protein ki-
nase (MAPK) cascade (5). SOS is activated by Ras-
GTP binding to an allosteric site, located between
the Cdc25 and Ras exchanger motif (REM) domains
in the catalytic core termed SOScat (C) (Fig. 1A)
(6). This allosteric activation depends sensitively
on the nucleotide state of Ras (7) and contributes
an important aspect of SOS biology. Function-
ally, this is thought to enable SOS to operate as
an analog-to-digital converter through a Ras-GTP
positive-feedback loop operating at the mem-
brane, such as during T cell activation (8, 9).
However, the nucleotide specificity of allosteric
activation of the catalytic site remains poorly
understood. SOS activation is autoinhibited by
its N-terminal Histone fold (H), Dbl homology
tween different activity states, is averaged out in
the ensemble result. Because many signaling pro-
cesses in cells involve small numbers of mole-
cules, the ability to average over large ensembles
is not a benefit that live cell-signaling networks
necessarily enjoy (14–17). Stochastic variation it-
self, rather than ensemble average properties, can
be a viable mechanism of regulation (18). To gain
a clear understanding of SOS activity and its
regulation on membranes will require direct ob-
servations of individual SOS molecules function-
ing in a membrane environment.
Single-molecule SOS activity assay
We developed an assay platform to observe real-
time single-molecule activity of SOS with Ras
in a partitioned, supported membrane (Fig. 1B).
H-Ras(C118S, 1 to 181) (referred to as Ras from
here on) was linked to the membrane through
maleimide coupling of Cys181 to supported lipid
bilayers (SLBs), resulting in stably bound, lat-
erally mobile Ras that is fully functional with
respect to SOS activation (19, 20). We also used
an H-Ras construct truncated at Cys184 [H-Ras
(C118S, 1 to 184)], which contains both native
cysteine palmitoylation sites (181 and 184) in the
hypervariable region (HVR), which we linked to
lipids through maleimide chemistry. Membranes
were composed primarily of L-a-phosphatidylcholine
(Egg, Chicken) (Egg-PC) with 2 to 3% anionic
1,2-dioleoyl-sn-glycero-3-phospho-L-serine lipids
(DOPS) and in some cases PIP2, which bind the
PH domain on SOS and release autoinhibition
(19). Molecules within the supported membrane
were confined in micrometer-scale corrals by lith-
ographically defined barriers to lateral mobility,
prefabricated onto the underlying substrate (21).
The barriers trapped membrane-tethered Ras
and SOS (through its interactions with Ras and
membrane lipids) within individual corrals. In
each corral, the membrane remained entirely

50 4 JULY 2014 • VOL 345 ISSUE 6192 sciencemag.org SCIENCE

 R ES E A RC H | R ES E A R C H A R T I C L E S

fluid, thus permitting SOS and Ras to diffuse surement of Ras lateral mobility with FCS and SOS binds to Ras on the membrane surface and,
freely while SOS processively catalyzed nucleo- fluorescence recovery after photobleaching (FRAP) in the absence of free nucleotide, becomes trapped
tide exchange. Ras surface density was calibrated (fig. S2) ensured that fluid bilayers were parti- (3). After free SOS was rinsed from the system,
with fluorescence correlation spectroscopy (FCS) tioned by leak-free corrals. unlabeled nucleotide was flowed in, and the ex-
(22) and subsequently measured in each corral SOS was introduced into the system from change reaction commenced. By loading Ras with
by means of epifluorescence imaging (fig. S1). solution in a transient pulse traveling through Atto488-labeled fluorescent nucleotide 2′/3′-O-(2-
Ras surface densities in these experiments ranged the flow cell. We examined several SOS con- aminoethyl-carbamoyl)-guanosine-5′-diphosphate
from several hundred to 1000 Ras molecules per structs derived from a truncated SOS containing or -[(b,g)-imido]triphosphate (Atto488-EDA-GDP
square micrometer, corresponding to the broad the H, DH, PH, and C domains but lacking the or Atto488-EDA-GppNp, respectively; referred
Ras density range measured in vivo (8, 19). Mea- C-terminal Grb2 binding domain (SOS-HDPC). to as GDP-488 and GTP-488, respectively, from
here on), catalytic exchange with nonfluorescent
nucleotide from solution (GDP or GTP) could be
directly observed as a local, corral-confined de-
crease of surface fluorescence. Constant flow during
the exchange reaction ensured removal of unbound
fluorescent nucleotides, allowing the reaction ki-
netics to be recorded by means of wide-field im-
aging of the fluorescence decay. Fluorescent and
nonhydrolysable nucleotide analogs did not sub-
stantially perturb SOS activity (20).
Representative images taken during SOScat-
catalyzed turnover showed a small percentage
of active corrals (those with a SOS molecule)
undergoing nucleotide exchange. These develop
a clear negative contrast relative to inactive cor-
rals (those without a SOS molecule) (Fig. 1C and
movie S1). In all experiments, SOS concentrations
were adjusted so that >95% of active corrals con-
tained exactly one enzyme (fig. S3) (20). Total in-
ternal reflection fluorescence microscopy (TIRFM)
of Atto647N-labeled SOScat (SOScat-647) revealed
a clear correspondence between active dark cor-
rals and individually confined SOScat enzymes
(Fig. 1D and movie S2), which were laterally mob-
ile within corrals (fig. S4) (20). However, it was
not necessary to label SOS to run the assay. By
imaging arrays of corrals on the surface, the ex-
change reactions of hundreds of individually con-
fined enzymes were monitored in parallel (Fig.
1E). Quantitative analysis of fluorescence decays
(fig. S5) (20) allowed individual kinetic traces of
SOS activity to be analyzed (Fig. 1F). The total
number of molecules tracked in an experimental
run was similar to that in some bulk experiments
Fig. 1. Platform for single-enzyme kinetics. (A) Schematic showing the crystal structure and domain (19), except that the individual contribution of
architecture of SOS-HDPC (surface and illustrated rendering) with Ras molecules (yellow, illustrated every single molecule was observed.
rendering) modeled onto the allosteric and catalytic sites, facing a lipid bilayer. SOS-HDPC is de- SOScat enzymes were highly processive on the
picted in a sterically closed and auto-inhibited conformation. The actual on-membrane configura- membrane surface. Hundreds to several thousand
tion is expected to be more open. The SOS-HDPC crystal structure, the HVR region of Ras, and the Ras could be activated by one SOScat molecule
1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC) bilayer structure are adapted from pub- during a single membrane residency period (Fig.
lished works (20). (B) Scheme of the experimental setup. Nanofabricated chromium metal lines 1G). Comparable results for SOS activity were
(10 nm high and 100 nm wide) partition a supported bilayer into micrometer-scale corrals, each observed with both the single- (C118S, 1 to 181)
containing lipid-anchored Ras loaded with fluorescent nucleotide at densities from hundreds to and double-anchored (C118S, 1 to 184) Ras con-
approximately a thousand molecules per square micrometer. When a single SOS engages Ras at structs (fig. S6). Because SOScat lacks any of the
the allosteric site, the catalytic site is free to turn over the remaining Ras in the corral, replacing other membrane binding domains, this confirms
fluorescent with nonfluorescent nucleotide and leading to a confined decrease of emission in- that Ras alone is sufficient to stably recruit SOS
tensity. (C) Wide-field epifluorescence image of fluorescently loaded Ras before injection of SOS to the membrane. A W729E point mutation in
(left) shows no initial dark corrals. Injection of a pulse of SOS followed by continuous flow of SOScat that abolishes Ras binding in the allo-
nonfluorescent nucleotide (right) leads to enzymatic turnover in a subset of corrals. (D) False color steric pocket (10, 19) inhibited processive activity
overlay of fluorescently loaded Ras emission (red) and TIRFM image of fluorescently labeled (fig. S7). Bivalent interaction with Ras through
SOScat-Atto647N (green) reveals colocalization of dark corrals and single SOScat enzymes. (E) Zoomed- both the catalytic and allosteric sites was required
out view showing a field of 1 × 1 mm2 corrals with SOS activity. (F) Collection of single-corral kinetic for sustained membrane localization of SOScat.
traces from an array of SOScat turning over Ras-GTP, showing the percentage of fluorescent Ras in the
corral as a function of time (blue traces). The black traces represent signals from empty corrals Dynamic heterogeneity of SOS activity
without enzymatic activity. (G) Histogram of total Ras-GTP-488 turnovers by individual SOScat Kinetic traces of nucleotide exchange from in-
enzymes. Scale bars, 10 mm (C) and (D); and 20 mm (E). Lipid composition (in molar percent): Egg-PC/ dividual SOS molecules revealed discrete tran-
MCC-PE/DOPS/TR-DHPE = 93.99/3/3/0.01. sitions between well-defined catalytic states.

SCIENCE sciencemag.org 4 JULY 2014 • VOL 345 ISSUE 6192 51

R ES E A RC H | R E S EA R C H A R T I C LE S

Traces from two representative SOS molecules,
each in its own corral, are plotted in Fig. 2A (more
traces are provided in fig. S8). One molecule ran
for more than 1200 s with a single catalytic rate
(1.1 molecules/s). The other molecule shown
started out with a similar rate, abruptly changed
to 4 molecules/s at ~400 s, then abruptly changed
back after another ~500 s. Using a change point
algorithm (20, 23) to detect transitions revealed
such events in 30% of traces for SOScat turning
over Ras-GDP (fig. S9). Changes in turnover rate
were not associated with changes in lateral dif-
fusion of SOS (fig. S10). Because the transitions
were discrete and the individual states exhib-
ited well-defined kinetic rates, we suggest that
these are the result of transitions between dif-
ferent stable configurations of the protein com-
plex itself (SOScat with Ras bound in its allosteric
site). Such conformational fluctuations have been
traditionally observed on the microsecond to mil-
lisecond time scale (24–26), but longer-lived dy-
namic conformers exist in proteins (27), extending
well into the minute time scale (supplementary
text S2).
By identifying individual functional substates,
we constructed the lifetime-weighted probabil-
ity histogram of distinct catalytic rates sampled
by hundreds of SOScat enzymes (Fig. 2B) (20).
The histogram spans almost 2 orders of magni-
tude, with a broad peak at ~1 s−1 and a wide
shoulder and tail of sparsely populated states
extending toward higher rates. The histogram
of apparent rates from inactive corrals, arising
from intrinsic nucleotide release and photo-
bleaching, represents the limit of resolution in
this experiment (20). Even in a single catalytic
state, stochastic variation resulting from the
discrete nucleotide exchange events will lead to
a real distribution of turnover rates over time
or PIP2 in these experiments) (10, 19, 28). How-
ever, we observed that the N-terminal domains
still influence SOS activity by allosterically sup-
pressing fluctuations that populate multiple
high-activity states, which is consistent with a
mechanism of conformational selection (supple-
mentary text S2) (29–31). This represents a pre-
viously unknown layer of SOS regulation and
emphasizes that not just the allosteric site, but
also the N-terminal domains, communicate di-
rectly with the catalytic site.
The Noonan syndrome associated R552G SOS
mutation leads to excessive activation of Ras
in vivo (13) and increases the apparent rate of
catalysis in bulk assays, but only when membranes
are present (19). In vesicle assays, Ras activation is
monitored as a function of the total amount of
SOS in the system, but the fraction of SOS actually
recruited to the membrane is unknown. The single-
molecule rate histogram for SOS-HDPC(R552G)
shows that the R552G mutation is not activating
at the level of SOS specific activity (Fig. 2C) and
results in a similar low-fluctuation frequency as
seen for SOS-HDPC (fig. S9). Taken together with
this observation, the accelerated turnover of Ras-
GTP by SOS-HDPC(R552G) reported from vesicle
assays (19) is likely to be a result of increased
surface binding from solution. Abnormally high
membrane recruitment of SOS-HDPC(R552G) may
contribute to the pathogenic hyperactivation of
Ras signaling in vivo as well.
Allosteric regulation of SOS
A second key observation is that the expected
activation of SOS by Ras-GTP bound in its allo-
0.0
ln(normalized int.)
steric site is surprisingly weak over most of the
range of the observed distribution of rates. Rate
histograms for SOScat, SOS-HDPC, and SOS-
HDPC(R552G) are shown in Fig. 3, A, B, and C,
respectively. Although small differences in the
distributions and most probable rates do exist,
it is unlikely that these minor effects would be
consequential in the context of a living cell. Some
cellular signaling processes can be triggered by
as little as a few individual ligand-receptor inter-
actions (15–17). In such cases, only a small num-
ber of SOS molecules are likely to be involved as
well. Variation of the ensemble mean (stochastic
noise) becomes large when sample sizes become
small, and thus, small differences in the mean
rates become obscured in stochastic noise (sup-
plementary text S4 and fig. S13).
However, the dynamics of SOS fluctuations
do exhibit differences, depending on whether
Ras-GTP or Ras-GDP is bound in the allosteric
site. Whereas SOS accesses the rare, highly active
states in either case, individual molecules linger
for longer periods in these active states when in-
teracting allosterically with Ras-GTP. This is made
vivid if we examine the total Ras activation by
individual SOS molecules. In Fig. 3, D, E, and F,
each observed catalytic state is plotted on the
basis of its specific activity and the Ras surface
density on which it was observed. The total Ras
turnover, equivalent to the number of Ras mol-
ecules activated by this particular SOS molecule
while in this state, is represented by the size
of the mark. Ras-GTP allosterically promotes
high-activity long-lived states at the expense
of low-activity shorter-lived states (figs. S14
1.0 SOScat
No enzyme

and between enzymes. We estimated this in-

Relative probability

Stochastic variation
trinsic stochastic noise through simulations of a
one-state system (supplementary text S3 and
fig. S11). The overall distribution of SOScat turn- -0.5 0.5
over rates is much broader than either the ex-
perimental or intrinsic stochastic noise limits.
However, it is likely that the assay does not re-
solve individual state peaks, giving instead the -1.0 0.0
appearance of a broad continuous distribution. 0 500 1000 1500 0 2 4 6 8

SOS regulation through Time (s) Rate (molecules/s)

intradomain interaction Fig. 2. Allosteric autoinhibition by N-terminal do-
0.4 SOScat
A key finding comes from a comparison of rate mains of SOS. (A) Logarithmic plots of single-enzyme
SOS-HDPC
histograms for SOScat and SOS-HDPC (Fig. 2C SOS-HDPC(R552G) kinetic traces. Discrete activity levels of SOScat are
0.3
and fig. S12). The N-terminal H, DH, and PH identified as linear segments. Some SOScat enzymes
Probability

domains have a pronounced effect on the dis-
tribution of rates sampled by SOS. Whereas the
peak of the rate distribution for SOS-HDPC
overlaps with that of SOScat, the extended tail
of molecules with faster rates is strongly sup-
pressed, resulting in a sharper histogram. The
N-terminal domains also dampen the frequency
with which state transitions occur, dropping to
10% of kinetic traces with transitions for SOS-
HDPC (fig. S9). These observations were made
under conditions in which autoinhibition was
previously thought to be released by interactions
between the PH domain and anionic lipids (PS
remain in a single functional state throughout the
0.2
reaction (straight trace, green), whereas others fluctu-
ate between different states (kinked trace, blue). The
0.1 black trace (square) represents corral that does not
undergo SOScat-catalyzed turnovers. (B) Normalized
0.0 and lifetime-weighted probability distributions of cata-
0 2 4 6 8 lytic rates sampled by (i) the SOScat ensemble (black),
Rate (molecules/s) (ii) corrals without enzyme (brown), and (iii) modeled
single-state enzyme stochastic variation (orange).
(C) Catalytic rate probability distributions from SOScat (black), SOS-HDPC (green), and SOS-HDPC
(R552G) (purple). The N-terminal domains auto-inhibit SOS specific activity. A Noonan syndrome–
associated point mutation in SOS-HDPC(R552G) (purple) does not relieve inhibition. Lipid composition
(in molar percent): Egg-PC/MCC-PE/DOPS/TR-DHPE = 93.99/3/3/0.01.

52 4 JULY 2014 • VOL 345 ISSUE 6192 sciencemag.org SCIENCE

 R ES E A RC H | R ES E A R C H A R T I C L E S

and S15). In cells, these highly active individ-
ual molecules would produce all of the Ras they
activate in the same location (supplementary
text S5) and over a short time. This is especially
consequential in a signaling context, in which
GTPase-activating proteins (GAPs) catalyze the
hydrolysis of GTP to GDP, thus inactivating Ras
at a basal level. The effects of such GAP activity
would render small bursts of Ras activation
inconsequential but may not keep pace with the
highly active molecules. In such a situation, the
average activity of the entire SOS ensemble is
relatively unimportant because the rare, highly
active individuals dominate the overall system
behavior. Furthermore, the important functional
consequences of feedback regulation of SOS by
Ras-GTP observed in computer simulations and
cellular experiments (8) could well be mediated
by the long-lived active states that allosteric reg-
ulation by Ras-GTP enables but that by Ras-GDP
does not. Ras reportedly forms dimers on mem-
brane surfaces (32). Such dimers could poten-
tially affect SOS activity and, depending on the
two-dimensional dimerization affinity on mem-
brane surfaces, could contribute to the mild
Ras density–dependence of SOS activity seen
in Fig. 3D.
Effect of fluctuation dynamics
on the signaling network
To explore the general feasibility of rate fluc-
tuations that provide a mechanism of allosteric
GDP -> GDP
10-1
GTP -> GTP
Probability
regulation, we conducted a series of stochastic
simulations, beginning with a simple enzymatic
system that does not account for feedback reg-
ulation (Fig. 4, A and B). The simple system
demonstrates the functional importance of rate
fluctuations and the existence of long-lived active
states. We consider a substrate, S, that is converted
to its active state, S*, by an enzyme E, which can
sample different possible states of activity (sup-
plementary text S6 and fig. S16). A deactivating
enzyme (analog of RasGAP) can convert S* to S.
We first carried out Gillespie simulations (33)
with single enzymes, with different trajectories
corresponding to different choices of the en-
zyme activity and switching rates. In one set of
trajectories, the enzymes were allowed to sample
only low-activity states (allosteric Ras-GDP–bound
SOS), whereas in another set, an additional high-
ly active state was sampled (allosteric Ras-GTP);
in another set of simulations, the enzymes sam-
pled identical enzymatic rate distributions but
transitioned between catalytic states at different
rates. Our simulations demonstrate that placing
additional weight in the highly active tail of the
distribution or having enzymes with long-lived
states can lead to fast threshold-crossing by a
subset of the enzymes, whereas a collection of
enzymes all operating at an increased but aver-
age catalytic rate are unable to support activation
(Fig. 4A). We also made simulations in which
the rates of a number of enzymes in a single re-
action group [such as a cell or a localized mem-
10-1 10-1
Probability
brane signaling cluster (34)] were chosen from
the same distribution, but their rates of switch-
ing between states were different. Histograms of
simulation results for many such reaction groups
showed that for small numbers of enzymes (ana-
log of weak stimulation), slower switching rates
can enable some groups to become activated,
whereas activity at the average rate would not
(Fig. 4B).
We also investigated a coarse-grained mod-
el of lymphocyte SOS signaling, which has been
used to interpret bimodal early signaling events
in lymphocytes (supplementary text S7) (8). This
model includes feedback regulation through
Ras-GTP and the interplay between SOS and the
RasGEF Ras guanyl nucleotide-releasing protein
(RasGRP), as well as the signal attenuation through
GAP activity (8). The resulting histograms of cel-
lular Ras-GTP levels are shown in Fig. 4C, each
containing thousands of simulated cells. The ex-
istence of stochastic activity fluctuations between
long-lived states in the simulated SOS-like GEF
shifted the regime of bimodal Ras-GTP signaling
(gray-shaded histograms) and broadened the
ability of the network to exhibit bistability. This
effect propagated through the network as a whole.
Bimodality resulting from static SOS (Fig. 4D,
red lines) or fluctuating SOS (Fig. 4D, blue lines)
differed across the SOS and RasGRP concentra-
tion matrix (Fig. 4D). Gray-shaded histograms
highlight regions of the parameter space where
only fluctuating SOS enzymes gave rise to bi-
modal Ras-GTP distributions, illustrating how
the threshold of the overall signaling network
is fine-tuned by stochastic rate fluctuations of
a single constituent enzyme—in this case, SOS.

Probability

10-2 Various models have been advanced to explain

10-2 10-2
10-3 how Ras-GTP (35, 36) and other cellular signals
10-4
10-3 10-3 (37, 38) are spatiotemporally shaped to process
10-5 10-4 10-4 information, for example, for analog or digital
0 5 10 15 20 0 5 10 0 5 10
response to stimuli. Stochastic rate fluctuations,
Rate (molecules/s) Rate (molecules/s) Rate (molecules/s) as reported here for SOS, constitute another
mechanism for tuning cellular signaling dynam-
20 10 10
ics (39).
Dynamic heterogeneity has been reported for
multiple enzymes (40–42). However, the dura-
Rate (molecules/s)

5
Rate (molecules/s)

10 5
Rate (molecules/s)

tion of dynamic SOS states we observed [several

minutes (Fig. 3B)] is longer than that reported
0 0 0 for other enzyme systems [up to tens of seconds
20 10 10
(41, 42)]. The SOScat fluctuation behavior shows
n n
n that dynamic conformational modes in signal-
12500 3400
2820
10 5 5 ing enzymes may extend well into time scales
7500 2040 1692 comparable with receptor signaling processes
2500 680
(37, 39). Evolutionary pressure is required to evolve
564
30 7 6
0 0 0
400 800 1200 400 800 1200 and maintain increased enzymatic efficiency (43),
400 800 1200
Ras density
so high-activity states are likely to be biologically
Ras density Ras density
(molecules/µm2) (molecules/µm2) (molecules/µm2) important. SOS is the key enzyme involved
in bimodal switching of Ras-GTP levels during
Fig. 3. Nucleotide specificity of allosteric activation of SOS. (A to C) The nucleotide-dependent rate T cell receptor–triggering in T cells (8, 44).
distributions for (A) SOScat, (B) SOS-HDPC, and (C) SOS-HDPC(R552G) on a logarithmic scale. For all It is certainly possible that rare high-activity
constructs, Ras-GTP bound in the allosteric site of SOS (blue traces) has a small activating effect relative states could function in threshold-crossing and
to Ras-GDP in the allosteric site (red traces). (D to F) Scatterplots of SOS turnover rate as a function of bistability, leading to robust Ras-GTP positive-
Ras surface density for (D) SOScat, (E) SOS-HDPC, and (F) SOS-HDPC(R552G). Each point in the scat- feedback activation of SOS and robust re-
terplots represents an observed catalytic state, and the area of the point represents the number of Ras sponse to antigen.
turned over by that catalytic state. For all constructs, Ras-GTP bound in the allosteric site of SOS (blue More fundamentally, we developed a single-
points) results in longer-lived high-activity states capable of highly processive catalysis. Lipid composition molecule enzymatic assay that enables detailed
(in molar percent): Egg-PC/MCC-PE/DOPS/TR-DHPE = 93.99/3/3/0.01. observation of the activation of membrane-linked

SCIENCE sciencemag.org 4 JULY 2014 • VOL 345 ISSUE 6192 53

R ES E A RC H | R E S EA R C H A R T I C LE S

16. B. N. Manz, B. L. Jackson, R. S. Petit, M. L. Dustin,

J. Groves, Proc. Natl. Acad. Sci. U.S.A. 108, 9089–9094
(2011).
17. G. P. O’Donoghue, R. M. Pielak, A. A. Smoligovets, J. J. Lin,
J. T. Groves, eLife 2, e00778 (2013).
18. P. J. Choi, L. Cai, K. Frieda, X. S. Xie, Science 322,
442–446 (2008).
19. J. Gureasko et al., Nat. Struct. Mol. Biol. 15, 452–461
(2008).
20. Materials and methods are available as supplementary
materials on Science Online.
21. J. T. Groves, N. Ulman, S. G. Boxer, Science 275, 651–653
(1997).
22. J. T. Groves, R. Parthasarathy, M. B. Forstner, Annu. Rev.
Biomed. Eng. 10, 311–338 (2008).
23. D. L. Ensign, V. S. Pande, J. Phys. Chem. B 114, 280–292
(2010).
24. D. D. Boehr, R. Nussinov, P. E. Wright, Nat. Chem. Biol. 5,
789–796 (2009).
25. K. Henzler-Wildman, D. Kern, Nature 450, 964–972
(2007).
26. K. A. Henzler-Wildman et al., Nature 450, 913–916
(2007).
27. W. Min, G. Luo, B. J. Cherayil, S. C. Kou, X. S. Xie, Phys. Rev.
Lett. 94, 198302 (2005).
28. C. Zhao, G. Du, K. Skowronek, M. A. Frohman, D. Bar-Sagi,
Nat. Cell Biol. 9, 706–712 (2007).
29. A. del Sol, C. J. Tsai, B. Ma, R. Nussinov, Structure 17,
1042–1050 (2009).
30. J. P. Changeux, Annu. Rev. Biophys. 41, 103–133
(2012).
Fig. 4. Modeling effects of stochastic transitions between long-lived enzymatic states by using a 31. P. Csermely, R. Palotai, R. Nussinov, Trends Biochem. Sci.
minimal signaling model or broader network model. (A) Fraction of trajectories (f) reaching a 35, 539–546 (2010).
32. J. Güldenhaupt et al., Biophys. J. 103, 1585–1593
threshold amount (NS*) of product S* during activation by single GEF-like enzymes fluctuating with
(2012).
switching frequencies (gS). GAP-like enzymes simultaneously deactivate S* to S at a fixed rate. The 33. D. T. Gillespie, J. Comput. Phys. 22, 403–434 (1976).
ensembles of enzymes represented by the three traces have identical mean activity levels and differ only 34. N. C. Hartman, J. T. Groves, Curr. Opin. Cell Biol. 23, 370–376
in the rate with which enzymes fluctuate between activity states (details of the model are provided in (2011).
35. A. S. Harding, J. F. Hancock, Trends Cell Biol. 18, 364–371
supplementary text S6). Enzymes with a fast switching rate (green trace) do not reach the signaling
(2008).
threshold, whereas enzymes with a lower switching rate (orange trace) do. Enzymes that do not switch 36. B. N. Kholodenko, J. F. Hancock, W. Kolch, Nat. Rev. Mol. Cell
between states on the time scale of the simulation show the most rapid activation and threshold Biol. 11, 414–426 (2010).
crossing. Additionally, no threshold crossing is observed for static enzymes operating at the average 37. H. Wu, Cell 153, 287–292 (2013).
38. H. E. Grecco, M. Schmick, P. I. Bastiaens, Cell 144, 897–909
catalytic rate. (B) Probability distributions of (NS*) for various ensemble sizes (Nenzyme = 4, 16, or 64),
(2011).
with slow switching (brown trace, gS = 10−3), fast switching (green trace, gS = 10−1), or enzymes 39. J. E. Purvis, G. Lahav, Cell 152, 945–956 (2013).
operating only at the average catalytic rate (blue trace). In all cases, the average catalytic rate is the 40. H. P. Lu, L. Xun, X. S. Xie, Science 282, 1877–1882
same. At low (4) and intermediate (16) copy number, the activation pattern depends strongly on the (1998).
switching rate. At low copy number, slow switching gives rise to a population of highly activating 41. B. P. English et al., Nat. Chem. Biol. 2, 87–94 (2006).
42. M. K. Prakash, R. A. Marcus, Proc. Natl. Acad. Sci. U.S.A. 104,
enzymes (resulting in more than 400 S*). (C) Distributions of Ras-GTP resulting from coarse-grained 15982–15987 (2007).
modeling of the Ras-SOS signaling network (details of the model are provided in supplementary text S7). 43. W. J. Albery, J. R. Knowles, Biochemistry 15, 5631–5640
Each histogram contains 8000 modeled cells. Left column shows simulations without catalytic rate (1976).
fluctuations, gS = 0. Middle column shows simulations with catalytic rate fluctuations for which the 44. J. P. Roose, M. Mollenauer, M. Ho, T. Kurosaki, A. Weiss,
Mol. Cell. Biol. 27, 2732–2745 (2007).
switching rate of SOS with Ras-GTP or Ras-GDP allosterically bound is the same: gGTP = gGDP. Right
column shows simulations in which SOS with Ras-GTP allosterically bound switches more slowly than AC KNOWLED GME NTS
did SOS with Ras-GDP allosterically bound: gGTP < gGDP. For each switching condition, the Ras-GTP Major support was provided by NIH P01 AI091580. Additional
distribution is plotted for increasing cellular levels of SOS. (D) Distributions of Ras-GTP resulting from support was provided by the Danish Council for Independent
static SOS (red traces) or fluctuating SOS (blue traces). The histograms span a matrix of cellular levels Research, Natural Sciences (L.I., S.M.C., and D.S.), by the
Mechanobiology Institute, National University of Singapore
of the two RasGEFs SOS and RasGRP. Conditions in which only fluctuating SOS gives rise to a bimodal
(C.-H.Y. and J.T.G.), and by the Berkeley Education Alliance for
Ras-GTP distribution are shaded gray. Research in Singapore (J.T.G.). L.I., H.-L.T., W.-C.L., and J.T.G.
planned and designed the research; H.-L.T. and L.I. performed
GTPases by GEFs. This system is generalizable 4. J. Cherfils, M. Zeghouf, Physiol. Rev. 93, 269–309 experiments and analyzed data; S.M.C. and W.-C.L. assisted with
(2013). experiments; H.-L.T., S.M.C., and H.-J.W. developed algorithms for
to a broad class of such interactions. We antici- data analysis; H.-L.T., S.M.A., and P.T. performed stochastic
5. A. Mor, M. R. Philips, Annu. Rev. Immunol. 24, 771–800
pate that fluctuations and distributions of rates (2006). simulations; J.I., J.G., and S.D.H. prepared proteins; C.-H.Y., C.R.,
may well prove to be just as important, if not more 6. S. M. Margarit et al., Cell 112, 685–695 (2003). and R.S.P. prepared patterned substrates; L.I., H.-L.T., and J.T.G.
so, than ensemble averages in the behavior of 7. S. Boykevisch et al., Curr. Biol. 16, 2173–2179 (2006). wrote the manuscript; J.T.G. supervised the project. All
8. J. Das et al., Cell 136, 337–351 (2009). authors commented on the manuscript.
cellular signaling networks. The single-molecule
9. J. E. Jun, I. Rubio, J. P. Roose, Front. Immunol. 4, 239
membrane surface assays introduced here render (2013).
such phenomena accessible to quantitative ex- 10. H. Sondermann et al., Cell 119, 393–405 (2004). SUPPLEMENTARY MATERIALS
perimental investigation. 11. K. K. Yadav, D. Bar-Sagi, Proc. Natl. Acad. Sci. U.S.A. 107, www.sciencemag.org/content/345/6192/50/suppl/DC1
3436–3440 (2010). Materials and Methods
12. J. Gureasko et al., Proc. Natl. Acad. Sci. U.S.A. 107, 3430–3435 Figs. S1 to S18
RE FE RENCES AND N OT ES
(2010). Tables S1 to S4
1. J. T. Groves, J. Kuriyan, Nat. Struct. Mol. Biol. 17, 659–665 13. M. Tartaglia et al., Nat. Genet. 39, 75–79 (2007).
References (45–69)
(2010). 14. E. Korobkova, T. Emonet, J. M. Vilar, T. S. Shimizu, P. Cluzel,
Movies S1 and S2
2. J. D. Scott, T. Pawson, Science 326, 1220–1224 (2009). Nature 428, 574–578 (2004).
3. J. L. Bos, H. Rehmann, A. Wittinghofer, Cell 129, 865–877 15. D. J. Irvine, M. A. Purbhoo, M. Krogsgaard, M. M. Davis, Nature 3 January 2014; accepted 10 June 2014
(2007). 419, 845–849 (2002). 10.1126/science.1250373

54 4 JULY 2014 • VOL 345 ISSUE 6192 sciencemag.org SCIENCE